If you’ve been following the conversation about technological unemployment and the threat of robots and A.I. stealing jobs, you may have come across the prediction that 47% of current jobs in the U.S. are at risk of automation. That figure comes from a widely cited 2013 paper, titled “The Future of Employment.”
One of that paper’s co-authors, Dr. Carl Benedikt Frey, has now expanded on the thesis in a new book. Frey is co-director of the Oxford Martin Programme on Technology and Employment at the U.K.’s prestigious Oxford University. His new book, The Technology Trap: Capital, Labor, and Power in the Age of Automation compares the age of artificial intelligence to past shifts in the labor market, such as the Industrial Revolution.
Frey spoke with Digital Trends about the impacts of automation, changing attitudes, and what — if anything — we can do about the coming robot takeover.
DT: There has been plenty of discussion of this topic over the past several years. What does your book add to the puzzle?
CF: There is a very polarized debate surrounding automation. The one extreme is that robots will take all the jobs, we’ll all be left unemployed, and the only solution is a basic guaranteed income. The other is people pointing to history and saying that automation has worked out well in the past.
I think that what the book does is to collect everything we know about the impact of automation. It gives an overview of the determinants of the pace of automation; taking in everything from the cost of capital relative to the cost of labor to attitudes to technological progress itself. Most importantly, it shows that everything didn’t always work out well for labor in the past. There were episodes when parts of the population faced declining wages for years — and even decades. And when people didn’t see technology improving their wages and conditions of living, they often opted against it.
Your prediction that 47% of jobs could be automated within the coming decades was widely reported. Do you feel that your concerns and conclusions in this paper have been accurately reported?
The paper received a wide variety of coverage. A lot of it has been good, but some of it has been less good. Overall, my impression is that few people actually read what we said in the paper. For example, we discuss how many determinants of technology adoption — like wages, legislation, culture, [and] resistance — can play into the pace of automation.
The paper also makes quite clear that the headline figure merely refers to the potential automatability of jobs from a technological capabilities point of view. It doesn’t say that these jobs will be automated or anything like that. I think it’s sometimes been taken to suggest that 47% of jobs will disappear in a decade or two. That was not what [myself and co-author Michael A. Osborne] said.
Do you think there is ever an argument for technological progress being halted on the basis that it will cause unemployment? In 1589, there’s a story about Queen Elizabeth I refusing a patent for a stocking frame knitting machine because it would put people out of work. In that scenario, permission was denied not on the inefficiency of the technology, but rather the impact that it would have. That seems difficult to imagine today.
My own view is that people who think we should bring progress to a halt haven’t really thought it through. If you were to stop the technological clock in 1900, that would clearly have been a mistake. People are much better off today as a result of technological change — both in their capacity as producers and consumers. I definitely think that progress over the long run is a good thing.
However, if you take the first Industrial Revolution, there were a lot of negative side effects for average people. Wages were stagnant, or even fell, for around seven decades. Not to mention the unhealthy working and living conditions in factory towns. The Luddites were essentially right to riot against the mechanized factory because they didn’t live to see its benefits from it. But future generations did. We can all be grateful that the Luddites didn’t succeed in bringing progress to a halt.
Are there certain jobs you think we as a society should be morally compelled to eliminate, even if that means putting people out of work? The modern equivalent of child chimney-sweeps in Victorian England.
I couldn’t give you one specific example of a job that we are morally obliged to automate. One of the more extraordinary things is how much of the hazardous work has already been eliminated, at least in the industrialized west. What we might deem hazardous work has fallen from around 60% to 10% over the past century. And much of the more routinized, boring work has disappeared as well.
In the developing world, there are still lots of unpleasant factory jobs which could be automated away. But they also support the livelihood of those people who hold them during a critical stage of development.
Are there jobs that you think are safe from automation not for reasons of technical bottlenecks, but because we as a society would not wish to hand them over to machines?
I think priests and politicians are two such examples. We’re unlikely to automate those for cultural reasons.
What has been the biggest surprise to you while researching this topic? Was there a trend you’ve observed, or a single piece of research, which has challenged your base assumptions on this topic?
What is most intriguing to me is to read through popular perceptions of technology in history. You find that the debates we are having haven’t actually progressed very much at all since the early eighteenth century, whereas the technology has progressed enormously. If you look at debates on automation in the 1930s or 1960s they’re extraordinarily similar to those we are having in the present day.
[Perhaps] the thing that surprised me the most is how much attitudes regarding what people think seem to matter for technology adoption. We assume that technology falls from the sky and we adopt it because it makes economic sense. But there are so many factors that play into this. One reason growth was so stagnant up until the Industrial Revolution — which could have happened much earlier because the technology was there — was that people didn’t see the introduction of replacing technologies as beneficial to them.
Craft guilds, in particular, vehemently resisted any technologies they perceived threatened their members skills. And fearing social unrest, governments often introduced legislation to block new technologies. Such was the political economy of technological change for most of human history.
Do you see sufficient areas of job growth today to offset the number of jobs being destroyed or negatively impacted?
I’m not concerned that we are not creating enough jobs. But I do think that we should be concerned about the fact that the wages of the unskilled have steadily fallen for three decades now. If we look at labor force participation rates, unskilled middle-aged men who used to work in the factories are now much less likely to have a job. I think this has a lot to do with the unevenness of job creation and job replacement.
If you think about the Bay Area, there are lots of new high tech industries. On the flipside, if you look at places like Detroit, a lot of the technologies which have been developed in the Bay Area have replaced people in Detroit. As a result of this, we see the local economy of Detroit has taken a hit. That’s because manufacturing jobs also supported the incomes of other people there as they went grocery shopping, took taxis, or went to the hairdresser. Meanwhile, when tech jobs are created in the Bay Area, that also creates more low-skilled service jobs in the area. This has led the great divergence we are seeing between skilled cities and the rest.
We’ve seen great correlation between technological advances and also a growing divide between rich and poor. Do you see causation here as well as correlation? Does technology necessitate this kind of hollowing out of the jobs market on one end of the spectrum and winner-takes-all hyper wealth on the other?
When it comes to the hollowing out of the middle of the labor market, there is an abundance of research which shows that automation and globalization have been the prime drivers. It’s hard to distinguish between the two because ICT been the enabler of globalization. Technological change and globalization has probably also to some to some extent driven the rise in top incomes, as it allows innovators and superstars in various fields to reach the global market places. But it also has a lot to do with compensation in the financial sector.
Another factor is housing. What is often overlooked is that almost the entire rise in wealth documented by Thomas Piketty has to do with housing. That, in turn, is related to structural changes in the economy. To return to the Detroit and Bay Area example, what happens is that when new tech jobs are created in the Bay Area, more people want to move in to tap into the local labor market.
That will drive up the cost of housing unless supply keeps pace with demand. Because of zoning restrictions, however, it rarely does. It also means that fewer people are able to benefit from the growth being created there.
People often talk about today’s technological revolution as fundamentally different to previous technological waves, particularly when it comes to employment. One reason for this is that we’re no longer simply seeing blue collar jobs replaced, but A.I. impacting on professional roles like lawyers and doctors as well. Do you view this as a difference?
I think it’s true that A.I. will also transform many skilled jobs. Medical diagnostics is one field that is already being automated. Certain tasks that lawyers used to do, like document review, is another example. But I think that doctors and lawyers have been relatively safe from automation because they also involve other tasks which are harder to automate, such as complex social interactions or creativity.
What our 2013 paper showed is that most skilled jobs are therefore not all that exposed to automation. The jobs that are much more exposed to A.I. are more in low-skilled sectors such as transportation, retail, logistics, construction. Although we will see A.I. moving into more professional services, I don’t think we will see much outright replacement there.
What advice do you have for people starting out in the workforce now, or trying to re-skill to ensure their future?
The good news is that the hardest things to automate are the things we enjoy, like social interactions and creativity. So it is not just about teaching digital skills. It is true that if you think that data is the new oil, then learning machine learning and statistics more broadly is a good idea. But I’m not a career advisor, and not an aspiring one either, so people are probably better off figuring out what they are good at themselves.
Initiatives like Universal Basic Income, robot taxes, and micropayments for data have all been put forward as ways to help protect workers in a future of automation. Are there any solutions you personally view as particularly viable?
I don’t think that there is one solution. But I do think that there are many things that we can do that, collectively, can make a big difference. Early childhood education is one such example. Deficits in math and reading, which emerge early in life, tend to be a bottleneck to further learning. People that lag behind early on are much less likely to go to college, meaning that is has significant impacts on their future earnings potential. Devoting resources to help people early on can make a really big difference.
If you think about the unevenness of job creation and replacement geographically, connecting places could also be very helpful going forward. Close to where I grew up in southern Sweden, Malmö was a city which had specialized in building ships. When that industry went down in the 1980s, Malmö declined. But it was given a boost by the construction of the Øresund Bridge between Malmö and Copenhagen in Denmark.
All of a sudden, people in Malmö had access to the labor market in Copenhagen. They could work there, but stay living in Malmö, where housing was relatively cheap, and spend their money locally, which boosted the local service economy. By connecting places that way, you can achieve quite a lot. There’s a currently a feasibility study looking at connecting Cleveland and Chicago using a Hyperloop. A six-hour commute would become 28 minutes, which would a feasible commute to work.
There are many other things that can be done which I discuss in greater detail in the book.
“The Technology Trap Capital, Labor, and Power in the Age of Automation” is published by Princeton University Press. This interview has been edited for length and clarity.
- Ford’s Dr. Derriere makes your car seats comfy, no ifs, ands, or butts about it
- Animals, algorithms, and obstacle courses: Welcome to the A.I. Olympics
- In the future, A.I. medicine will let patients own their health data
- What is Google Duplex? The smartest chatbot ever, explained
- Can A.I. truly be creative? One brilliant Brit makes a compelling case